Pen Pointing At Brain Targetted Medicine

Machine learning: The next frontier of drug discovery and diagnosis

Posted on 02 March 2021 by Wayne Marshall

Pen Pointing At Brain Targetted Medicine

Machine learning: The next frontier of drug discovery and diagnosis

Posted on 02 March 2021 by Wayne Marshall

x

​The drug development path is long, complicated, and expensive. For the patients that need it most, waiting for a treatment can be agonising, and currently 95% of rare diseases lack any FDA-approved treatment. Artificial intelligence and its subfield of machine learning has the power to change this.

Approval rate for drugs entering clinical development is less than 12%, and with the average cost of developing a prescription treatment sitting at $2.6 billion, those in drug development are turning to technology that detects signs that might indicate a greater likelihood of product success or approval, whilst the product is still in the discovery or R&D phase.

It can take over 10 years from discovery to commercialisation of a treatment, so a huge motivator for large pharma to invest in this type of technology is the potential to shave time from this process. An oncology programme from Genentech (now a member of the Roche Group) and GNS Healthcare announced back in 2017 uses machine learning tech to covert high volumes of cancer patient data into computer models that can be used to identify new pathways, novel targets and diagnostic markers that could lead to new treatments.

The benefits of utilising artificial intelligence reach far beyond just a cost or time saver for the company developing the drug. Synchronicity of real-world data and machine learning technology can shorten the amount of time it takes for a patient to receive a diagnosis and reduce the likelihood of a misdiagnosis. In medical diagnosis a patient’s symptoms are recorded, analysed, and a healthcare professional aims to determine the ailment that could be causing them. Machine learning technology can take a set of data – consisting of clinical or biological criterion of patients, any diagnoses they’ve received, and any environmental and genetic data – and match these with the data collected from a patient. The larger the data set, the more points of reference the AI will have to determine a diagnosis, and the greater the likelihood of an accurate one.

One example of a firm currently using machine learning to aid diagnosis is UK-based start-up Babylon Health. Their chatbot technology compares the symptoms inputted by a user against a database of potential causes, and recommends an action based on the symptom severity, the history of the patient and other individual circumstances. In response to mild headache symptoms, the app may recommend over-the-counter medication, but if more serious symptoms are inputted, the app may suggest going directly to the hospital.

This does present a limitation in that the performance and accuracy of the technology is dependent on the depth of the dataset, but it’s a promising indicator that these processes could take place without the need for human intervention, meaning patients get quicker access to the care and information they need.

Machine learning is becoming more widespread, is attracting a huge amount of media attention and capital investments, and outcomes are becoming increasingly impressive. There is still a distance to cover, though, before it really hits the mainstream. Before cementing itself as a standard component in the drug development cycle, educational programmes, thorough proof-of-concept studies and data hygiene standards have to be considered. Until then, these early applications of machine learning in medicine provide glimmers of a future where data and technical innovation play a critical role in discovery and development.